Research
Notes on Choosing a Rubric Judge
Experiments in rubric grading
Kos-1 Experimental: Env-Free RL on a 1T Parameter Agentic Prior
Scaling medical RL to Kimi K2.5.
Kos-1 Lite: SOTA Medical Model
Introducing Kos-1 Lite, our state-of-the-art medical model.
The LLM Data Company x Perplexity: DRACO
Our work with Perplexity to create frontier evaluations for Advanced Deep Research.
Mismatch Praxis: Rollout Settings and IS Corrections
Notes on solving training/inference mismatch.